v0.3.0

Full Changelog: v0.2.0...v0.3.0

v0.3.0 - 2026-05-05

Added

• Added agent command transcript logging: opt-in config, append-only NDJSON storage, CLI stdout/stderr tee capture, managed-shell capture, task transcript command, and task report --include command-log section. Export/import preserves transcript archives.
• Added planning guidance profiles: plan guidance command, --include-guidance plan template injection, and prompt_profiles.planning config with advisory required-fields rendering.
• Added transcript review mode as the default task transcript output, with --raw flag for the original table view, duplicate log ID warnings, and logical-row grouping for wrapper/managed-shell pairs.
• Added enriched command metadata: tier, deprecated, replaced_by, ledger_effect, workspace_effect, external_effect, and agent_safe fields on CommandSpec. Added --tier and --include-deprecated CLI filters. Deprecated lock break in favor of repair lock.
• Added first-class expired-lock resume path: implement resume --repair-expired-lock releases expired implementation locks with audit trail, and next-action emits expired-lock-resume when applicable.
• Added task-resource positional refs for read-only commands (task show, task view, etc.) with explicit --task required for destructive commands (task cancel, task uncancel, task edit).
• Added JSON usage-error envelopes for workflow positional-ref rejection and CLI parse errors.
• Added soft task archive: task archive/task unarchive commands, archived-task visibility filtering across list/tree/status, and slug reuse semantics.
• Added export/import project metadata guard, dry-run safety, and include-flag controls.
• Added plan revision workflow: plan export, plan amend, and --auto-revise with safe plan input path guard and plan.amended audit events.
• Added plan approval provenance: approval_source and approved_plan_hash stored on acceptance, with hash-mismatch warnings in reports.
• Added implement finish warning for missing git change scans.

Changed

• Split run/lock helpers into services/run_store.py from the tasks.py monolith; tasks.py re-exports for backward compatibility.
• Wrapper commands now mirror inner exit status by default instead of always succeeding.
• Planning guidance recommendation is now integrated into plan start and next-action with a one-time viewed marker.
• Plan lint human output now renders summary and issue details instead of bare pass/fail.
• Question answer-many now validates repeat inputs, aliases, and provenance.
• Doctor mismatch guidance and verbose output improved with actionable repair hints.

Fixed

• Fixed task report Plans section so non-accepted plans show reviewable details instead of being omitted.
• Fixed pre-commit --all-files regressions across test files.

Documentation

• Documented planning guidance profiles in README, usage, command contract, API, and skill.
• Documented transcript logging, managed command capture, and review mode in usage and skill.
• Documented expired-lock-resume path and --repair-expired-lock in SKILL.md.
• Documented command-surface safety guidance, task-resource positional refs, and destructive-target rules in SKILL.md and command contract.
• Documented plan revision workflow commands and safety semantics in SKILL.md and command examples.
• Updated failure-review remediation hints for known mistakes in docs and skill.

Quality

• Added regression test modules for agent command logging, expired-lock resume, task archive, and plan revision workflow.
• Expanded command inventory, CLI contract, JSON contract, and docs/skill tests for metadata enrichment, deprecation, targeting, and envelope behavior.
• Full suite: 770+ tests passing, ruff and mypy clean.